AI draws a beautiful poster, but the QR code scans into nothing. AI writes a page, but you can't describe which heading you mean. That's the one thing PodApp does.
Windows 10 / 11 · Installs per-user, no admin rights · Unsigned installer — SmartScreen will warn; click “More info → Run anyway”
Ready the moment you install. Clicking the icon, calling it headless from AI, and calling it over MCP all run the same code path.
The dock clings to the right edge of your screen; if the Codex / ChatGPT desktop app is open it snaps beside that window and follows it. Alt + Space brings it up anywhere. Drop an image on it to start.
.pod and it becomes an MCP tool automaticallyEvery action has a stable ID and runs headless. The button a human clicks and the call an agent makes go through the same function — not two implementations. Write it twice and the first requirement change forks them, and after the fork the UI still looks right while the AI path quietly breaks.
podapp action run app.nine-grid.image.split --json \
--input '{"image":"poster.png","rows":3,"cols":3,"zip":true}'
A pod is a manifest, a web UI, and a set of actions. The spec, JSON Schema, zero-dependency CLI and the AI skill pack are all open source: podapp-protocol.
npx podapp create my-pod
npx podapp pack my-pod